Product photo sizes and aspect ratios: one image, several channels

Every place you post has a fixed shape it puts images into. A product grid needs every cell to match, so it favours squares. A feed you scroll with your thumb favours tall images, because they take up more of the screen. Short clips and stories fill a phone screen, which is very tall. When your photo is one shape and the frame at the other end is another, the system there decides for you what to cut off and where to leave empty space — and it has no idea where in the picture your product is.

The ratio is the shape of the frame: as wide as it is tall, or taller than it is wide. Pixels are how much detail sits inside that frame. Those are two different questions with two different fixes, and the ratio comes first, because the ratio is what decides whether your product gets cut off — and it is the part you can actually set in our app before you press Generate. We offer four shapes: 1:1, 4:5, 3:4 and 9:16.

• 1:1 is a square, as wide as it is tall. It is the shape product grids commonly use, because every cell in a grid has to be identical.

• 4:5 is a shallow portrait, slightly taller than it is wide. It claims a lot of feed space without being tall enough to risk having its top and bottom trimmed.

• 3:4 is one step taller again. A full-length garment fits comfortably, and it is the shape the system picks for you when you ask for a full-length shot without choosing a ratio.

• 9:16 is the tall shape that fills a phone screen. It is the shape of short clips and stories, and a still image in this shape sits behind a clip edge to edge, with no black bars above and below it.

• There is no landscape option at all. All four choices are square or portrait. That is a real limitation, and the reason for it is in the next section.

• Try-on · has the Ratio control.

• Product to Model · has the Ratio control.

• Ghost mannequin · has the Ratio control.

• Image Set · no control. The system sets a shape per image based on that image's framing. A set contains both full-length and half-length shots, so the whole set comes out portrait, but not all in the same shape.

• Upscale · no control, because it does not draw a new picture. Whatever shape went in comes back out, larger.

• Background removal · no control, for the same reason. It cuts the background out of the existing picture and the frame stays exactly as it was on every side.

The first option under Ratio is called Let the system choose, and it is the default. That phrase used to mean we sent no value at all and let the provider guess, which did not work. The project's owner fired 8 test images in one day and got 4 different shapes back. So we counted every old job in the database and found that jobs sent without a ratio came back not-portrait somewhere between roughly one in five and almost half the time, even though the instruction asked for a full-length shot — while every job that was sent a ratio came back in the shape it was asked for. Since 2 September 2026, Let the system choose means the system always sends a shape that matches the framing. It is no longer silence.

• Full-length framing: the system sends 3:4.

• Half-length framing: the system sends 4:5.

• Close-up framing: the system sends 1:1.

• Ghost mannequin has no framing control, so with no ratio chosen it comes out 3:4.

• Try-on is the single exception. If you choose nothing, we send nothing, because we measured 11 jobs with no value sent and every one came back the same shape. That provider already has a firm default of its own, and overriding it would be a risk with no problem behind it.

This is the most important sentence on the page. Cropping removes what you do not want, and it can never create a side of the picture that is not in the file. That is why the order in which you choose the shape matters so much, and the arithmetic behind it is simple. When you crop a portrait image to a square, the tool keeps the full width and trims the height. What you lose is the top and the bottom — the head and the hem — not the left and right sides.

• From 4:5 to square: about one fifth of the height is cut. This is the cheapest crop there is.

• From 3:4 to square: about a quarter of the height goes. Half-length shots survive it; full-length ones usually lose the feet or the hem.

• From 9:16 to square: nearly half the height goes. A full-length shot almost never survives.

• From square to 9:16 is worse still, because nearly half the width has to go, leaving a narrow central strip. If the product is not dead centre, it gets clipped.

• If you have to pick one shape as the source for every channel, 4:5 is the most flexible, because it crops down to square with the smallest loss of height. A 9:16 version is better generated separately than cropped out of another shape.

• The crop tool only offers Full image, 1:1, 3:4 and 9:16 as ready-made frames — there is no 4:5 button yet. If you want 4:5, choose it at generation time, or drag a free frame and judge the proportion by eye, which is hard to get exact. That gap is ours, and we would rather write it down than hide it.

The ratio says what shape the image is. The pixels say how much detail is inside that shape. Cropping always costs you the second one, because every edge you cut away is fewer pixels left. Crop hard enough that the shorter side falls below 480 pixels and our system will not take that image back in to work on. Pressing Upscale after a crop does not bring back detail you have already cut off, either. If you know in advance that you will be cropping several shapes out of one image, premium quality on Product to Model produces a larger file than standard, which leaves you more pixels to cut into.

One more symptom that looks like a ratio problem but is not: you shoot a photo upright, and somewhere else it appears on its side. Most phones store the picture rotated and write a tag inside the file saying how far to turn it before displaying. Software that ignores that tag shows the picture as it was actually stored — sideways. When you upload to us, we physically rotate the file upright according to that tag before storing it, and then strip every piece of hidden data out, including the GPS location phones attach. Images that leave our system are already the right way up and do not depend on that tag any more.

There is not a single platform's pixel figure on this page, and that is deliberate. We searched twice and found nothing usable. What we found was other companies' blog posts, an old document for a different country's marketplace, and a page written for developers — none of them a help page for Thai sellers today. Those numbers also change whenever the platform feels like it, without telling us. If we print one and it changes, you are the one who loses time. The most reliable figure is the one on the help page of the place you are actually listing on. Open it there and follow it there. What we can guarantee are our own requirements, because we wrote them.

• Choosing a ratio before you generate costs no extra credits, whichever shape you pick.

• Cropping with the Edit image button costs nothing either, because it runs on your device and fires no AI job.

• What costs credits is generating a new image: Product to Model 2 credits per image at standard quality and 5 at premium, Try-on 2 credits per image, ghost mannequin 2 credits per image at standard quality and 4 at premium, upscale 1 credit per image, background removal 1 credit per image.

• Re-running the same image just to change its shape is charged in full every time, exactly like a new image. That is the entire reason this page insists on choosing the shape first and cropping afterwards.

• Failed jobs are refunded automatically without you having to ask, but an image that generated successfully in a shape you did not want is not refunded.

Signing up gives you 10 free credits, once, with no card required. With no extras selected that is 5 Product to Model images at standard quality — enough to check whether the shape you picked comes out the way you pictured it. One thing to know before you spend them: your first image on the free plan has no watermark, but every image after it carries our watermark burned into the file itself, not laid over the screen, and it is a pattern repeated across the whole image rather than one mark in a corner, so cropping does not remove it. The free plan is there for judging quality, not for doing real work.